Microsoft Fabric Licensing: F-SKU Capacity, Pricing & Negotiation

What Microsoft Fabric Is and Why Licensing Changed Everything

Microsoft Fabric, launched at Build 2023 and reaching general availability in November 2023, is Microsoft's unified data platform — consolidating Azure Data Factory, Azure Synapse Analytics, Azure Data Lake Storage, Power BI, and a new set of data engineering, science, and real-time analytics capabilities under a single SaaS product. The unification matters commercially because it replaced several independently licensed Azure services with a single capacity-based SKU model — Fabric Capacity, denominated in Compute Units (CUs) and sold in F-SKU tiers.

For enterprises that previously licensed Power BI Premium, Azure Synapse, and Azure Data Factory independently, Fabric's launch created both opportunity (simplified licensing, potential consolidation savings) and risk (the migration path is not automatic, and Microsoft's Fabric pricing is structured to grow your spend rather than reduce it). Our Microsoft advisory team has reviewed Fabric proposals across financial services, manufacturing, and retail enterprises and finds that the majority of initial Fabric quotes are priced 30–45% above what organisations with comparable workloads achieve through negotiated capacity commitments. Explore the full context in our Microsoft Knowledge Hub.

F-SKU Capacity Tiers: Understanding What You're Actually Buying

Microsoft Fabric capacity is sold in F-SKU tiers: F2 (2 CUs), F4, F8, F16, F32, F64, F128, F256, F512, F1024, and F2048. The CU is the fundamental billing unit — it determines how much computational work your Fabric capacity can process simultaneously. Different Fabric workloads consume CUs at very different rates: a Power BI report with complex DirectQuery against a large semantic model, a Spark notebook processing a 10TB data lake file, and a real-time analytics stream processing 50,000 events per second all consume CUs differently and compete for the same shared capacity pool.

The practical consequence is that F-SKU sizing is genuinely difficult without actual workload data — and Microsoft's pre-sales sizing is systematically optimistic. The most common pattern we see is: an enterprise migrates from Power BI Premium P2 to Fabric F64 (Microsoft's suggested equivalent), discovers that Fabric workloads run concurrently rather than queuing the way Premium Gen2 did, exhausts the F64 capacity during business hours, and is pushed toward F128 within the first quarter. Before committing to any F-SKU, pressure-test Microsoft's CU consumption estimates against your specific workload patterns. The relationship to Azure cost optimisation is also important — Fabric's OneLake storage is separately billed on Azure storage rates and is not included in the F-SKU.

Pay-As-You-Go vs Reserved Capacity: The Commercial Trade-Off

Microsoft Fabric capacity can be purchased in two modes: pay-as-you-go (PAYG), billed per CU-hour at a premium rate; and reserved capacity, committed monthly or annually at a significant discount. PAYG provides flexibility but is prohibitively expensive for sustained workloads — the per-CU-hour rate for PAYG is typically 3–4x the effective rate of a one-year reserved commitment. For any enterprise planning sustained data platform usage, reserved capacity is the correct commercial model. The negotiation question is not whether to reserve, but at what SKU level, for what term, and whether Azure committed use discounts apply to Fabric capacity in addition to Azure services.

Fabric capacity reserved pricing through the Azure portal is list price — the same rate available to any customer with a credit card. Negotiated EA pricing for Fabric capacity is available and meaningfully different from list, but only through direct EA negotiation, not through the Azure portal. Enterprises that purchase Fabric capacity as a standalone Azure reservation without involving their EA team are systematically overpaying. The interaction with Azure Reserved Instance commitments is also worth understanding — Microsoft's account teams sometimes present Azure commitments and Fabric reservations as separate negotiations, but they should be positioned together to maximise leverage.

Migrating from Power BI Premium and Synapse: Cost Impact

Power BI Premium P-SKU customers migrating to Fabric have two options: migrate to F-SKUs (the standard Fabric capacity model), or retain Power BI Premium P-SKUs which Microsoft continues to sell but is clearly sunsetting in favour of Fabric. The F-SKU migration path offers Fabric's full capability set including data engineering, data science, and real-time analytics — but P-SKU to F-SKU migrations frequently reveal that the "equivalent" F-SKU Microsoft recommends is undersized for concurrent workloads that were queued under Premium Gen2.

Azure Synapse Analytics customers migrating to Fabric face a more complex transition: Synapse pipelines move to Fabric Data Factory, Synapse Spark pools move to Fabric Spark, and Synapse dedicated SQL pools have no direct Fabric equivalent — requiring either migration to Fabric Warehouse (different performance characteristics and billing model) or continued running of the Synapse SQL pool as a separate Azure service alongside Fabric. This hybrid architecture is common and creates a period of parallel licensing cost that Microsoft's migration guides significantly understate. For organisations also implementing GitHub Copilot for the development teams building data pipelines, the combined Microsoft data and developer tooling cost deserves unified commercial review. Book a call to model your specific migration economics.

Copilot in Fabric: The AI Add-On Pricing Trap

Microsoft is actively adding Copilot capabilities to Fabric — natural language querying of data models, AI-generated notebook code, automated data pipeline suggestions, and anomaly detection in real-time analytics streams. Copilot in Fabric is included in F64 and above SKUs at no additional licence cost, but requires the capacity to be in a region where Copilot is available (currently limited) and consumes a portion of the capacity's CU budget for AI inference workloads.

The practical risk is that activating Copilot in Fabric on an already-busy F64 capacity can push overall CU consumption over the threshold — creating an artificial forcing function to upgrade to F128 that is driven by AI workload overhead rather than genuine data processing demand. Before activating Copilot in Fabric, benchmark CU headroom on your current capacity and model the incremental consumption before enabling it for broad user populations. Our guide on Microsoft AI licensing covers the full Copilot portfolio economics, including where Fabric's AI capabilities sit relative to M365 Copilot and GitHub Copilot in terms of per-user value.